Samsung Finally Has Foundry Pricing Power, and It Is Charging China For It

Samsung Electronics has raised prices on some new foundry orders by as much as 15%, a move Reuters reported on the 18th and tied to AI chip demand pushing TSMC's advanced-process capacity to saturation ^{[1] [2]}. For a contract chipmaking unit estimated to have lost money since 2022, the number matters less than the direction: Samsung is setting terms rather than discounting to fill lines ^[4].

The increases are specific rather than blanket. According to sources cited in the report, Samsung raised prices in July on new orders for its 4-nanometer SF4 process, with U.S. and Chinese customers paying 10-15% more than the previous month and Taiwanese customers 5-10% more ^[5]. SF5 wafers on the 5-nanometer node also went up 10-15%, and the older 8-nanometer process rose close to 10% ^[6]. That last one is the tell: 8nm is not where AI accelerators are built, so the pricing power is spreading beyond the nodes that are actually short.

The geography of the increase is not accidental. Reuters said Chinese customers are absorbing the steepest rises, attributing it to U.S. export controls on advanced chipmaking equipment that make domestic advanced capacity hard to build, leaving Chinese firms dependent on foreign foundries ^[7]. Samsung, meanwhile, cannot take everything, since it also has to reserve capacity for U.S. customers and for its own chips ^[10]. Samsung declined to comment in detail on the report ^[11].

Scale keeps this honest. Counterpoint puts TSMC above 70% of global foundry revenue in the first quarter against roughly 7% for Samsung, a gap of about ten to one ^{[3] [12]}. A share shift measured in single points does not change that ranking, but it can change Samsung's income statement. Lee Min-hee of BNK Investment & Securities said customers unable to secure TSMC volume are moving to Samsung and Intel, and that further price increases could push the foundry business into profit as early as the start of next year, sooner than expected ^[8].

The utilisation picture supports the pricing. A person familiar with the company said the SF4 line at Samsung's Pyeongtaek campus has run at maximum utilisation since late last year, producing both external logic chips for customers such as Qualcomm and base dies for Samsung's own high-bandwidth memory ^[9]. That is the useful mechanism here: HBM is a core component of AI accelerators, so the same demand wave is filling the memory business and the logic lines underneath it ^[13]. Samsung said in July that the foundry would return to profit soon on better utilisation, improved yields and steadier prices, and guided to second-half foundry revenue growth of more than double digits year on year ^[14]. It expects advanced processes to exceed half of foundry revenue this year, with AI and HPC products rising from 15-20% late last year to more than 30%, close to a doubling of mix in a year ^{[17] [18]}.

Watch three things. Whether the July increases stick into fourth-quarter contract renewals, or whether they were opportunistic pricing that resets when TSMC adds capacity. Whether the customer pipeline converts: Tesla and Apple announced Samsung production contracts last year, and the report says talks now extend to Broadcom, Nvidia and Google ^{[15] [16]}. And whether the China mix becomes a liability, given that the customers paying the most are the ones most exposed to export-control policy ^{[7] [10]}.